To guide behaviour, it has been proposed that neurons eventually learn to predict future states of sensory inputs. The project mentors have worked in this direction to get metrics on these predictions about how accurate those predictions are if the neuron used synaptic learning rules. The main contribution of this project would be to publish highly optimised library codes that can serve as evaluation benchmarks for predictive accuracy. We also believe that neurons can generate efficient encodings on these predictions. Through this project, estimates of the predictive information would also be obtained by neural models.



Shiven Tripathi


  • Sarah Marzen
  • cav